Objectives  Global navigation satellite system (GNSS) is an essential tool for landslide monitoring. On the one hand, influenced by unmodeled errors such as multipath, some abnormal fluctuations will occur in the monitoring sequence of GNSS, which will have a negative impact on the accurate discrimination of hazard warning, and even cause serious consequences such as missed warning and false warning. On the other hand, the state of landslide hazard bodies is predictable throughout their life cycle, such as the commonly used three phase law: Initial acceleration stage, constant velocity stage and instability acceleration stage. However, this feature information is not fully utilized in the GNSS solution process.

  Methods  For the above problems, the state space model of GNSS hazard monitoring is analyzed, and a new real-time filtering algorithm is proposed considering the state characteristics of the hazard body. The algorithm models the historical state information of monitoring points through the adaptive adjustment of the size of the sliding window, and then reasonably adjusts the current state parameters, so as to obtain a more reliable real-time solution sequence. The monitoring data of Heifangtai loess landslide in Gansu Province are selected for experimental verification.

  Results  The experimental results show that compared with the conventional results, the monitoring accuracy of the new real-time filtering algorithm for floating point solution and fixed solution can be improved by 97.6%, 87.5% and 89.6% in the east, north and up directions, respectively. The monitoring accuracy of fixed solution can be improved by 50.0%, 14.3% and 18.8%. The fixed ambiguity rate can be increased from 97.1% to 99.9%.

  Conclusions  The new real-time filtering algorithm not only improves the accuracy of monitoring results, but also improves the fixed rate of the ambiguity, which effectively reduces the influence of abnormal sequence fluctuation on hazard monitoring and warning.